Seminar 2 - Algorithms comparison

We can see that the # of classes is balanced - 40 of each

After looking at the histograms, we can see that some algorithms have more instances where they better performance (e.g., Alg 12 has 30 data points where the meand_ind was over 0.75, whereas Alg 1 has only 22)

However, some of these performances could've been obtained by using better parameter values.

We will look into that next.

We can see that param_[1, 2, 3] always have the same values (i.e., [10, 0, 5]). Only param_4 has different values so we will use it to make a plot.

Depending on the dataset, Alg10 has different performances. It deals really poorly with dataset of id 8, but it's best on dataset 5. In these cases, we can see that the value of param_4 does not affect the performance, only slightly. Let's also look at Alg1.

Again, only param4 varies.

We again see that the value of param_4 has only a small impact on the results. The algorithm however performs differently depending on the dataset.

Next, we are going to look at the ind_0, ..., ind_9 metrics. Since there are 10 of them, they will be hard to plot, thus, running PCA seems like a good idea.